-
Notifications
You must be signed in to change notification settings - Fork 89
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Use mlir_mode::fast for dot operations on Navi and reduce k threshold to 1024 #2608
Conversation
This looks like this is missing the k<1024 change. |
find_mlir_standalone_convolution_op{get_mode("convolution", mlir_mode::int8)}, | ||
find_mlir_standalone_dot_op{get_mode("dot", mlir_mode::none)}); | ||
find_mlir_standalone_convolution_op{get_mode("convolution", mlir_mode::fast)}, | ||
find_mlir_standalone_dot_op{is_navi ? mlir_mode::fast : get_mode("dot", mlir_mode::none)}); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This should probably be get_mode("dot", mode)
so we use MLIR for dot fusions when using the MIGRAPHX_ENABLE_EXTRA_MLIR
setting.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Sorry thats not the case. This is standalone.
Can this be targeted to develop and then cherry picked into the release branch? |
This can be closed. We won't be using this technique |
There will not be a need to target this against rocm-rel-6.0. If you have other uses in 6.1 then go for it but a new PR might be easier |
No description provided.